Multimodal

This work proposes VRCLIP, a multimodal person ReID framework that fuses RGB and RF signals by reformulating cross-modal alignment as correlation maximization via canonical correlation analysis. Featuring an illumination-disentangled encoder and RF-anchored adaptive fusion, VRCLIP achieves 93.9% mAP on the newly introduced large-scale VRR dataset, demonstrating robust performance under challenging lighting and occlusion conditions.